NIH Public Access: Author Manuscript

NIH Public Access
Author Manuscript
Vision Res. Author manuscript; available in PMC 2008 September 1.
Published in final edited form as:
NIH-PA Author Manuscript
Vision Res. 2007 September ; 47(19): 2499–2505.
Letter case and text legibility in normal and low vision
Aries Arditi and Jianna Cho

Arlene R. Gordon Research Institute, Lighthouse International, 111 East 59th Street, New York, NY
10022, Email: aarditi@lighthouse.org, Voice: 212 821 9500, Fax: 212 751 9667
Abstract
It is thought by cognitive scientists and typographers alike, that lower-case text is more legible than
upper-case. Yet lower-case letters are, on average, smaller in height and width than upper-case
characters, which suggests an upper-case advantage. Using a single unaltered font and all upper-, all
lower-, and mixed-case text, we assessed size thresholds for words and random strings, and reading
speeds for text with normal and visually impaired participants. Lower-case thresholds were roughly
0.1 log unit higher than upper-. Reading speeds were higher for upper- than for mixed-case text at
sizes twice acuity size; at larger sizes, the upper-case advantage disappeared. Results suggest that
upper-case is more legible than the other case styles, especially for visually-impaired readers, because
smaller letter sizes can be used than with the other case styles, with no diminution of legibility.
INTRODUCTION
There is conventional wisdom, supported by some evidence and logic within the fields of
typography and cognitive science, that asserts that text set in mixed upper- and lower-case is
more legible than all upper-case (all capital letters). Typographers generally point to the fact
that word shape is more distinctive with mixed- and lower-case than it is with all upper-case,
a virtue that results from the fact that all upper-case characters are the same height and have
no ascenders and descenders, whereas lower case characters, which have both ascenders and
descenders vary in both height and average position, arguably making words constructed with
them more distinctive due to more variation in the height of word contours (see Figure 1).
Miles Tinker, an authority on legibility and typography said “Lower-case letters have more
‘character’ in terms of variation in shape and the contrasting of ascenders and descenders with
short letters. This leads to characteristic word forms that are much easier to read than words
in all capitals” (M. Tinker, 1963 p. 34) Tinker found that while upper-case text was perceived
at a greater distance, it had a ‘retarding effect’ on reading speed, especially for long intervals
of reading, and was preferred by only 10% of readers, compared with 90% for lower-case text.
(M. A. Tinker, 1932; M. A. Tinker & Patterson, 1929)
The evidence from cognitive science comes from tachistoscopic experiments that suggest that
letter identification follows word identification rather than preceding it. Cattell (1886) early
on showed that with tachistiscopic presentation, words are recognized more accurately than
letters, a phenomenon that in various guises and variations, has come to be known as the “word
superiority effect.” While there are alternate explanations of this and related effects, such as
the greater ease with which letters are recognized within words than in isolation, it has been
taken as evidence for a dominant role of word shape in word recognition, relative to letter
recognition. Since lower-case words appear to have more distinct shapes than upper-case, there
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting
proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could
affect the content, and all legal disclaimers that apply to the journal pertain.
Arditi and Cho Page 2
is the common belief that the word superiority effect is responsible for what is assumed to be
the greater legibility of lower-case text.
There is a very sensible competing idea, however: that all upper-case text should be more
legible since the letters are in general larger than in lower-case text. Enlarging nearly any small
object makes it more visible, of course, and indeed nearly all optical vision aids rely on the
enhanced visibility of magnified objects to achieve better visibility. Text set in visually small
sizes in general, and low vision reading in particular, might be expected to benefit from the
larger letter sizes of upper-case letters.
The issue of size complicates matters considerably in studies of legibiliity, however, since there
does not seem to be any method for characterizing letter size that properly accounts for both
upper- and lowercase letters. One may equate uppercase and lowercase character size by
equating cap height (the height of a capital letter) to the x-height of lowercase letters . This
generally results in findings of lowercase being more legibile than uppercase (Smith, Lott, &
Cronnell, 1969). Since 12 of the 26 lowercase letters have ascenders and descenders that extend
well above the x-height or below the baseline, respectively, this method of specifying letter
size gives an unfair size advantage to lowercase letters. Characterizing letter size by overall
font size (conventionally equal to cap height plus descent, at least for computer fonts), similarly,
gives a size advantage to uppercase letters since 14 of the lowercase letters (those without
ascenders or descenders) are smaller than virtually all the uppercase letters, in both breadth
and height.
In this paper, we adopt the latter convention, of specifying letter size by font size, i.e. by the
sum of cap height plus descent, which is usually specified in points. We do so because font
size is usually specified in this fashion in typography and graphic design, without distinguishing
between upper- and lowercase character size. This decision will allow us to make make very
practical conclusions that can be applied by any graphic designer. In the discussion, of course,
we will consider the inherent size difference between letter cases, and weigh the potential
advantage of enhanced word shape information in lowercase words against the letter size
advantage of words set in upper-case.
METHODS
We assessed relative legibility of different case conditions using three different criteria for
legibility:
1. Size thresholds (visual acuity) for letter identification, measured with 5-letter, strings
presented on a video monitor, using an up-down staircase (Levitt, 1971) with 0.05
log unit size steps. Size (or, inversely, distance) thresholds are probably the most
common method for assessing text legibility (M. Tinker, 1963), and are widely used
in applied settings such as highway signage, with lower size thresholds indicating
higher legibility. We used two kinds of stimuli: random strings of all lower-, all-upper,
and randomly selected case and 5-letter words, all upper- or all lower-case, randomly
selected from the 2110 most frequent 5-letter words in English (Francis & Kucera,
1982).
2. Reading speeds using rapid serial visual presentation (RSVP). Higher legibility, by
this criterion, allows faster reading. We measured reading speed using RSVP with
small (two times acuity size) and large letters (roughly ten times acuity size), using
both all upper-case and conventional mixed-case text from an expanded MNREAD
(Legge, Ross, Luebker, & LaMay, 1989) corpus. Reading speed is a less common
measure of legibility but it is perhaps more representative of ordinary reading than is
size threshold. And because RSVP can support extremely high rates of reading (Rubin

& Turano, 1992), it has the potential to be more sensitive to subtle differences in
legibility. RSVP reading was tested with individual sentences, whose speed was
varied to determine the speed that supported a 50% correct (of words) reading rate.
3. Reading speeds using continuous reading of text passages taken from standardized
tests (9th grade level). We included this condition to address possible differences
between reading speeds with RSVP with those more commonly observed with
continuous reading.
We also characterized participants’ degree of vision loss with by assessing visual acuity with
a transilluminated Lighthouse/ETDRS distance acuity chart. These measurements were also
used in the computation of acuity reserve (see below).
Stimuli
Size thresholds (Experiment 1)—In this experiment, random 5-letter strings were
presented centered on a SONY Multiscan 520GS monitor, as black (3.6 cd/m2) letters on a
white (129 cd/m2) background. Normally-sighted subjects viewed the screen optically folded
through a front-surface mirror at an optical distance of 788.4 cm, so that letters were at least
100 pixels in height (from the top of an upper-case letter to bottom of the descent), or
equivalently, for these fonts, 66.66 pixels in cap height. For these subjects, the letters were
rendered in reverse on the screen to compensate for the mirror reversal. Subjects with low
vision viewed the screen directly (i.e. with no mirror) at a viewing distance of 100 cm.
Participants were seated comfortably in a chair, with head position fixed with a head and chin
rest.
The random letter strings were constructed by sampling (with replacement) from the 26 letters
of the English alphabet, and for the random case condition, then selecting the upper- or
lowercase version of the letter randomly with probability 0.5.
RSVP reading (Experiment 2)—We used custom software to present each word of a
sentence centered vertically and horizontally on the computer monitor, for a constant time
interval. Text was black on white, as with the size thresholds. The participant read aloud each
sentence as it was presented, prior to presentation of the next sentence.
Continuous reading (Experiment 3)—Four text passages of ninth grade-level reading

difficulty, and approximately 400 words in length, were used. The subject read the text aloud
continuously, while the experimenter timed the reading of the entire passage and recorded
errors.
Font—We used TrueType Arial as the display font for the entire study. Arial was selected
because it is found on most computers used for desktop publishing today, and because it has
a large x-height, making it relatively less likely to produce legibility differences based on
differences in relative size of upper and lower case letters. Font point size for the reading speed
measurements (Experiments 2 and 3) was set to an acuity reserve (Whittaker & Lovie-Kitchin,
1993) of 2, such that the x-height of the lower-case letters would subtend twice the visual angle
of the letters of the visual acuity chart. Additional measurements (Figure 5) were made with
acuity reserve of 10 (ten times the size of the chart letters).
Participants
Normally-sighted participants were Lighthouse research staff (two participants), or recruited
from the Lighthouse International Volunteer Service (two participants). Low vision
participants were recruited through the Lighthouse Low Vision Service, and were identified
for potential participation in the study by search of the Lighthouse Consumer Information

System. Participant details are shown in Table 1, which also shows the specific experimental
conditions each participant was tested in. All low vision subjects had clear ocular media and
their visual acuity loss was due to macular dsyfunction. All participants, both normal and low
vision, were naïve to the purposes of the experiment.
Experiment 1: Size thresholds

Procedure—Size thresholds were measured using a staircase method (Levitt, 1971) in which
correct identification of at least 4 of 5 letters (in correct order) was required for a decrease in
letter size on the subsequent trial, while no more than three letters correct elicited a size increase
on the subsequent trial. For a 26-letter stimulus set size, this procedure converges on the 68.6
percent correct point on the psychometric function.
Subjects were required to give 5 letter responses to all trials, and were encouraged to guess if
they reported difficulty. On trials in which the size changed, the magnitude of the change was
0.05 log unit, half the size change from line to line on state-of-the-art visual acuity charts. Data
prior to the 2nd reversal of each staircase were discarded, in order to concentrate the data used
in the analysis close to the threshold.
For each subject, data were collected for random letter strings first, and words subsequently.
Each case condition was run four times, in an order that did not favor any condition with respect
to practice (see Table 2). Each run terminated after 15 staircase reversals. Since each condition
was run four times, each condition’s threshold was estimated from 52 staircase reversals (13
reversals from each run).
All responses were given verbally by the subject; the experimenter typed the responses into
the computer, which then presented the next 5-letter string whose size was contingent on the
subject’s performance. Letters were presented continuously until subjects responded. Subjects
were able to change their responses if they did so prior to the experimenter’s finalizing the
response to that line. This procedure results in a negligible lapse, or extraneous noise rate
(Arditi, 2006).
Results and Discussion

Log size thresholds (in arc min of visual angle) are shown as a function of letter case for three
normally-sighted participants in Figure 2 and the four participants with low vision in Figure
2. The thresholds for each subject are geometric means of all the staircase levels visited (after
the second reversal of each run); the number of measurements on which the thresholds were
based ranged from 73 to 116. Standard errors (s.e.’s) about these means (which reflect accuracy
of values in terms of proportion, rather than magnitude) were small; the maximum ratio of s.e.
to threshold over all participants and all stimulus conditions was 0.041.
Repeated measures linear mixed effects modeling (Pinheiro & Bates, 2000) revealed that for
both normally-sighted and low vision participants (whose size thresholds differed, F1,6 = 58.82,
p < 0.0003), thresholds are lower for words than random letter strings (F2,25 = 344.23, p <
0.0001), evidencing a word superiority effect. In addition, thresholds for all caps conditions
were lower than those for lower case (F2,25 = 12.19, p = 0.0002) or mixed case conditions. The
mixed case condition fell intermediate to the UPPER and lower case conditions, for normally-
sighted participants but not those with low vision, whose thresholds were highest in the mixed
case condition.
One might expect that the mixed-case condition would fall intermediate to the upper- and
lower-case conditions, as they did for the normally-sighted participants, simply because on
average, the stimuli were of intermediate size. However, the task is also more difficult than
either the upper- or lower-case conditions because the a priori probability of a correct guess

is reduced from 1/26 to 1/52. In the present experiment, note also that the variability in the
threshold estimates was considerably larger for the low vision participants than the normally-
sighted participants, making our finding of the highest thresholds in the mixed-case condition
for low vision participants less certain.
Experiment 2: Reading speed

In this experiment, we assessed the affect of letter case on reading speed using both the RSVP
reading technique and continuous text presentation. Continuous presentation was of ‘pages’
composed of lines of text displayed on a monitor. This kind of reading is not unlike reading
from a book or periodical, except that viewing distance and display luminance are better
controlled.
Our rationale for using RSVP, in which words are presented one at a time in the center of the
computer monitor, was that since it allows reading at higher speeds than with continuous verbal
reading (Rubin & Turano, 1994;Rubin & Turano, 1992), especially for normally sighted
readers (Rubin & Turano, 1992). RSVP, then, might plausibly be more sensitive to subtle
differences in legibility. In order to further enhance this sensitivity, we used sentences from
an expanded MNREAD corpus. These sentences are by design 56 characters long (including
interior spaces) with roughly comparable comprehensibility (See Mansfield, Ahn, Legge, &
Luebkerr, 1993 for details). Since the sentences are very short, readers can store most or sall
of each sentence in short-term memory, and report the sentence verbally without needing to
maintain a high rate of verbal output, which might otherwise limit speeds.
Procedure
RSVP—Word presentation rate, which was controlled by a desktop computer, was varied only
between sentences, by an amount that was contingent on reading error rate. Because we had a
limited number of 56-character MNREAD sentences (357), and wanted to obtain error rates
for a range of presentation rates. Subjects were given practice on 60-character MNREAD
sentences prior to testing. The experimenter determined informally during the practice phase
the speed region in which the subject began to make errors, by increasing speed by 20% if no
errors were made, and decreasing speed if errors were made. Once data collection began, the
speed increments and decrements were reduced to 10%, and data collection proceeded in
staircase fashion, such that if no errors were made, the speed was increased (by the
experimenter); if no words in the sentence were correctly identified, the speed was reduced.
Our goal was thus to obtain nonzero error rates for several presentation speeds, sampling a
range of the sloping portion of the psychometric function. We obtained estimates of between
5 and 10 speeds for each of the two case conditions (upper and mixed case), for each subject.
Error rates (in characters per 56-character sentence) were then fit by fit by probit (Finney,
1971), to a cumulative Gaussian. Maximum reading speed was taken to be the speed in words
per minute, at which 50% errors were made. Following the method of Carver (1976), speeds
in words per minute were computed by assuming that each sentence was composed of 9.33
standard length words (each six letters in length) and dividing by the exposure time for the
sentences.
It might be argued that the cognitive load of remembering the text strings might differentially
affect younger and older subjects; however, we cannot test this since since all of our younger
subjects were normally-sighted and only one older subject had normal vision. But the
comparison of interest here is letter case within normally-sighted and low vision group, who
obviously differ significantly in overall reading speed.
Continuous text—In the continuous reading speed measurements, simple text passages of
ninth grade reading level and roughly 400 words in length, were presented on the screen with

text wrapped and no hyphenation, one screenful at a time, with subsequent screens elicited by
the experimenter, until the passage read was completed. The experimenter recorded errors, and
the reading speed recorded was the number of words correctly read divided by the reading time
in minutes.
Viewing distances and font sizes for both the RSVP and continuous reading measurements,
were chosen to approximate an acuity reserve of about 2 for all participants, which is close to
the maximum reserve our low vision observers had available to them due to their relatively
poor reading acuity (Lovie-Kitchin & Whittaker, 2000). Subsequently, we assessed reading
speed in two normally-sighted readers at much larger letter sizes (see Results and Discussion
below).
Results and Discussion

50% correct RSVP speed thresholds and error-corrected continuous text reading speeds for the
three normally-sighted participants, are shown in Figure 3. Data from the same conditions are
shown for the four low-vision participants in Figure 4. For both of measures, and both types
of subjects, upper-case text produced faster reading speeds (F1,20 = 5.530, p = 0.029), again
supporting the idea that all upper-case text is more legible than mixed-case text.
Note that while our low vision readers were reading with roughly the same acuity reserve as
they had when reading under non-experimental conditions, the same was not true for the
normally-sighted readers, who typically read at many multiples of their threshold letter size.
As an afterthought, we decided to assess reading speed at a visual size corresponding to more
typical reading conditions for normally-sighted readers. We chose an acuity reserve of 10
because Whittaker & Lovie-Kitchin’s (1993) analysis suggested that this resulted in the highest
reading speeds for normal readers. The data, shown in Figure 5, suggest that the upper-case
advantage disappears, when text is large enough. This is also consistent with Legge et al.’s
(1985) finding that there is a relative plateau in the function relating reading speed to print size
in the mid- to large-print size range; that is, there is a substantial range of print sizes above the
critical print size, at which reading speed changes very gradually. Since size itself has little
effect on reading speed in this region, small differences in size associated with letter case will
also have little or no measurable effect. Note, however, that we used a font (Arial) with a
relatively large x-height, which would tend to minimize differences based on relative size of
upper- and lower-case letters.
CONCLUSION
Our finding that size thresholds for upper-case text were lower than those for lower-case text
in Experiment 1 are not surprising, and corroborate the findings of (M. Tinker, 1963) that at
great viewing distance (as simulated by small visual size), upper-case text is more legible, even
in a font with a relatively large x-height, which might be expected to minimize upper- and
lower-case differences. Other fonts, which typically have smaller x-heights, might be expected
to show upper-case text to have even greater relative legibility. Contrary to Tinker’s findings,
and the conventional wisdom, is the result that upper-case text is more legible in terms of
reading speed, for readers with reduced acuity due to visual impairment, and in normally-
sighted readers when text is visually small. This result may have practical significance as well;
it suggests that, apart from economic considerations of how much space a given sample of text
occupies, letter size determines legibility for low vision readers and for those viewing visually
small text; and when point size is fixed, upper-case text is simply more legible, albeit less
aesthetically appealing, than lower-case.

Acknowledgements
We thank Gordon Legge, Steve Mansfield, Beth O’Brien and Lee Zimmerman for the expanded MNREAD corpus
used in this study. The work was partially supported by NIH grants EY12465, AG14586, and EY015192, and grants
from the Hoffritz and Pearle Vision Foundations.
References
Arditi A. Lapses in the verbal reporting task. Vision Research 2006;46(8–9):1327–1330. [PubMed:
16043209]
Carver RP. Word length, prose difficulty, and reading rate. Journal of Reading Behavior 1976;8:193–
203.
Cattell JM. The time taken up by cerebral operations. Mind 1886;11:277–282. 524–538.
Finney, DJ. Probit Analysis. 3. Cambridge: Cambridge University Press; 1971.
Francis, WN.; Kucera, H. Frequency analysis of English Usage Lexicon and Grammar. Boston:
Houghton-Mifflin; 1982.
Legge GE, Pelli DG, Rubin GS, Schleske MM. Psychophysics of reading. I. Normal vision. Vision
Research 1985;25:239–252. [PubMed: 4013091]
Legge GE, Ross JA, Luebker A, LaMay JM. Psychophysics of reading. VIII. The Minnesota Low-Vision
Reading Test. Optometry and Vision Science 1989;66(12):843–853. [PubMed: 2626251]
Levitt H. Transformed up-down methods in psychoacoustics. Journal of the Acoustical Society of
America 1971;49:466–477.
Lovie-Kitchin J, Whittaker SG. Prescribing near magnification for low vision patients. Clinical &
Experimental Optometry 2000;82:214–224. [PubMed: 12482267]
Mansfield JS, Ahn SJ, Legge GE, Luebkerr A. A new reading-acuity chart for normal and low vision.
OSA Technical Digest 1993;3:232–5.
Pinheiro, JC.; Bates, DM. Mixed-effects models in S and S-Plus. New York: Springer Verlag; 2000.
Rubin GS, Turano K. Low vision reading with sequential word presentation. Vision Research 1994;34
(13):1723–1733. [PubMed: 7941378]
Rubin GS, Turano KA. Reading without saccadic eye movements. Vision Research 1992;32(5):895–902.
[PubMed: 1604858]
Smith F, Lott D, Cronnell B. The effect of type size and case alteration on word identification. American
Journal of psychology 1969;82:248–253. [PubMed: 5811185]
Tinker, M. Legibility of print. Ames, Iowa: Iowa State University Press; 1963.
Tinker MA. The influence of form of type on the perception of words. Journal of Applied Psychology
1932;16(4):167–174.
Tinker MA, Patterson DG. Influence of type form on speed of reading. Journal of Applied Psychology
1929;13(6):205–219.
Whittaker SG, Lovie-Kitchin J. Visual requirements for reading. Optometry and Vision Science 1993;70
(1):54–65. [PubMed: 8430009]

Figure 1.
Shapes of words in (a), outlined in (b), are more distinctive in mixed-case and lower case than
all upper-case text.


Figure 2.
Means of normally-sighted participants' (n = 3) log size thresholds (in arc min visual angle)
for identification of words and random strings as a function of letter case condition. Error bars
indicate +/− 1 s.e.m.

Figure 3.
Means of log size threholds of participants with low vision (n = 4) as in Figure 2.

Figure 4.
Mean reading speeds for MNREAD sentences presented by RSVP (left) and continuous text
(right) for the two case conditions for normally-sighted readers (acuity reserve = 2; n=3).

Figure 5.
Average reading speeds for MNREAD sentences presented by RSVP (left) and continuous text
(right) for the two case conditions for low vision readers (acuity reserve = 2; n=4).

Figure 6.
Average reading speeds for MNREAD sentences presented by RSVP (left) and continuous text
(right) for the two case conditions for two normally-sighted readers (acuity reserve = 10; n=2).

NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript
Table 1
Characteristics and experimental conditions run, for the 9 participants of the study.
subject age diagnosis log MAR Snellen equivalent experiment
CC 28 normal vision 0.0 20/20 size threshold reading
speed (a.r. = 10)
KB 30 normal vision −0.2 20/13 reading speed (a.r. = 2)
RA 20 normal vision 0..0 20/20 size threshold reading
speed (a.r. = 2)
RL 34 normal vision −0.1 20/16 reading speed (a.r. = 2)
Arditi and Cho
LH 72 normal vision 0.2 20/32 size threshold reading

speed (a.r.= 10)
IR 72 diab.retinopathy 0.9 20/159 size threshold, reading speed (a.r.= 2)
LG 83 ARM 1.2 20/317 size threshold reading
speed (a.r.= 2)
MG 76 ARM 1.1 20/252 size threshold reading
speed (a.r.= 2)
SM 77 ARM 0.9 20/159 size threshold reading
speed (a.r.= 2)

Page 14
Table 2
Order of condition blocks within runs for random letters (nonense) and word identification conditions in
Experiment 1.
Run Random letters Words
1 All cap, all low, mixed All cap, all low
2 Mixed, all low, all cap All low, all cap

3 Mixed all low, all cap All low all cap
4 All cap, all low, mixed All cap, all low

NIH Public Access: Author Manuscript

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

NIH Public Access: Author Manuscript

Uploaded by

Copyright:

Available Formats

NIH Public Access

Vision Res. 2007 September ; 47(19): 2499–2505.

Letter case and text legibility in normal and low vision

Aries Arditi and Jianna Cho

Vision Res. Author manuscript; available in PMC 2008 September 1.

Continuous reading (Experiment 3)—Four text passages of ninth grade-level reading

Vision Res. Author manuscript; available in PMC 2008 September 1.

vision, were naïve to the purposes of the experiment.

Experiment 1: Size thresholds

Results and Discussion

Vision Res. Author manuscript; available in PMC 2008 September 1.

for low vision participants less certain.

Experiment 2: Reading speed

Vision Res. Author manuscript; available in PMC 2008 September 1.

Results and Discussion

Vision Res. Author manuscript; available in PMC 2008 September 1.

from the Hoffritz and Pearle Vision Foundations.

Vision Res. Author manuscript; available in PMC 2008 September 1.

all upper-case text.

Vision Res. Author manuscript; available in PMC 2008 September 1.

Vision Res. Author manuscript; available in PMC 2008 September 1.

Vision Res. Author manuscript; available in PMC 2008 September 1.

Vision Res. Author manuscript; available in PMC 2008 September 1.

Vision Res. Author manuscript; available in PMC 2008 September 1.

Vision Res. Author manuscript; available in PMC 2008 September 1.

LH 72 normal vision 0.2 20/32 size threshold reading

Vision Res. Author manuscript; available in PMC 2008 September 1.

2 Mixed, all low, all cap All low, all cap

Vision Res. Author manuscript; available in PMC 2008 September 1.

You might also like